Research On Early Fault Diagnosis Method Of Variable Speed Rolling Bearing Based On Time-frequency Analysi

As a key component of various nuclear power equipment,the failure of rolling bearings often leads to the functional failure of the entire nuclear power equipment or even the entire nuclear power system.Therefore,early diagnosis and research of bearing faults have important engineering significance.In response to the long-term variable speed operation of bearings under actual working conditions,and the traditional algorithms that rely on speed measurement devices,slow operation speed,and low accuracy when processing time-varying non-stationary signals of bearings,this paper focuses on time-frequency analysis technology and studies the feature extraction and intelligent classification methods of variable speed bearing faults under strong background noise.The main research content is as follows:(1)Due to the time-varying fault feature frequency and fixed fault feature order caused by early bearing faults under variable speed operating conditions,using Computational Order Tracking(COT)technology is a common variable speed fault feature extraction algorithm,but it is difficult to effectively extract the signal fault feature order in actual operating conditions using this technology alone.To this end,a method based on the combination of adaptive frequency modulation mode decomposition and COT is proposed.Firstly,the original signal is decomposed and processed,and each component is calculated using the square envelope Gini coefficient and square envelope spectral Gini coefficient indicators.Based on the calculation results,the optimal component is selected,and the fault feature order of the optimal component is obtained through the COT algorithm.(2)In order to eliminate the dependence on speed devices and interpolation techniques,this paper proposes a method based on a combination of adaptive short time fractional Fourier transform(ASTFrFT)and multi-source time-frequency ridge extraction.Firstly,the ASTFrFT algorithm is used to obtain the time-frequency map of the signal,and then the multi-source time-frequency ridge extraction method and the strategy of average fault feature coefficients are used to extract the fault feature order of the variable speed bearing.(3)In order to overcome the drawbacks of slow computational speed,low accuracy,and weak generalization ability of classical deep learning models in early fault intelligent classification of variable speed bearings,this paper proposes an intelligent classification and recognition method based on the combination of ASTFrFT and time-frequency BoTNet models.Firstly,various signals are divided into equal numbers of samples in a batch processing manner,and the ASTFrFT method is used to represent each sample in time frequency;Construct a time-frequency BoTNet model and initialize it,input the divided training and validation set samples into the model for training and validation;Finally,input the test set samples into the trained model for testing,and output the classification results.The data used in this paper include simulation signals of variable speed bearing faults and the University of Ottawa bearing experimental dataset.The feature extraction section was validated through simulation and experimental signals,while the intelligent classification section was validated through experimental signals.The effectiveness and superiority of the proposed method were verified by comparison with other methods.At the end of this paper,a summary of the work done,main innovative points,and future work is provided.